End-To-End Memory Networks for Contextual Language Understanding

2. A system as claim 1 recites, wherein the neural network includes a recurrent neural network (RNN).

3. A system as claim 1 recites, wherein the operations further comprise calculating the attention distribution as a softmax of the inner product of the embedding of the current input with individual embeddings of the historic inputs.

4. The system of claim 1, wherein the historic inputs comprise inputs from one or more users other than a user that provided the current input, the inputs being weighted based on a strength of association between the one or more users that provided the inputs and the user that provided the current input.

5. The system of claim 4, wherein the strength of association between the one or more users that provided the input and the user that provided the current input is based on an association between family members, an association between coworkers, geographical proximity between the users, or shared context between the users.

6. The system of claim 5, wherein the strength of association is based on shared context between the users, the shared context comprising at least one of demographic context, geolocation context, context from a knowledge graph, context from a search-engine log, or language context.

8. A method as claim 7 recites, wherein the neural network comprises a recurrent neural network (RNN).

9. A method as claim 7 recites, wherein the attention distribution is calculated as a softmax of the inner product of the embedding of the current input with the embeddings of the historic inputs from the store of inputs.

11. A system as claim 10 recites, wherein the neural network model includes a recurrent neural network (RNN).